径向基函数网络泛化能力研究及其应用

Research on the generalization of radial basic function neural network and its application

通过分析径向基函数网络(RBF网络)结构及影响其泛化能力的因素,提出一种通过网络泛化误差自动调节隐层节点数以得到最佳节点数的方法。将采用这种算法的RBF网络用于对船舶焊接变形过程进行建模并预测其输出。仿真表明这种算法可以显著提高传统RBF网络的泛化能力。

A novel algorithm of radial basic function neural network （RBF network） is presented based on the analysis of the structure of RBF network and the factors affecting the generalization, so that an improved RBF network is achieved, with the optimum number of neurons on the hidden layer. Then a modified network is adopted in modeling and prediction process of ship-welding deformation. The simulations indicate that generalization ability of the RBF network can be improved a lot.